The Consumer Credit Channel of Monetary Policy* Wenlan Luo† September 2017 Abstract Using the Consumer Expenditure Survey, I document a new fact that the con- sumption response to monetary policy shocks is greater for households with higher default risk. I propose a consumer credit channel that accommodative monetary pol- icy extends credit disproportionately to risky households which have higher propen- sities to spend out of extended credit. I study the mechanism in a Heterogeneous Agent New Keynesian model augmented with asymmetric information. In the model, credit limits arise because borrowers can default on loans and borrowing signals a risky type. Accommodative monetary policy extends credit as it lowers default rate and changes lenders’ beliefs on the types of borrowers. Calibrated to match the cross-sectional distribution of default rate, credit limit, and marginal propensity to spend, the consumer credit channel accounts for 63% of the heterogeneous consump- tion responses and 20% of the aggregate response. The model is used to assess the distributional effects of monetary policy and the "risk-taking" channel. Keywords: consumer credit, monetary policy, default risk, adverse selection, het- erogeneous agents. *I thank Mark Huggett for his guidance, inspiration and support. I thank Adrien Auclert, Jinhui Bai, Dan Cao, Olivier Coibion, Behzad Diba, Jingting Fan, Pedro Gete, Andy Glover, Kurt Mitman, Emi Nakamura, Jon Steinsson, Franco Zecchetto for helpful comments. All errors are my own. †Tsinghua University. Email: [email protected]. Website: http://luowenlan.weebly.com 1 1 Introduction A satisfactory understanding of the monetary transmission mechanism is the basis of any effective conduct of monetary policy. The current paper focuses on the transmission mechanism through the largest component of GDP: household consumption. The tradi- tional New Keynesian model emphasizes the demand response to changes in the policy rate from a representative consumer, which are inconsistent with two basic facts. On the one hand, macro evidence suggests that the consumption sensitivity to changes in the interest rate is small (Campbell and Mankiw (1989)). On the other hand, micro evidence shows that there exists substantial heterogeneity in borrowing limits, and households facing tight borrowing limits respond strongly to the changes in quantity but not the price of available credit (Gross and Souleles (2002)). This points to a gap in the literature that the response of credit supply to different households may be important in driving consumption response but is absent in traditional models. The current paper fills in this gap. Using the Consumer Expenditure Survey, Ifirst document a new fact that the consumption response to monetary policy shocks is greater for households with higher default risk. The results can be best summarized in Figure 11. After a negative innovation in the federal funds rate, the consumption impulse re- sponse for households with higher default risk is two times larger than the average, while it is virtually zero for those with lower default risk. I study a model featuring en- dogenous credit extensions in both prices and quantities that explains the heterogenous consumption responses. I use the model to answer the question: what is the role of con- sumer credit extensions in driving heterogeneous and aggregate consumption responses to monetary policy shocks? The model builds on the Bewley-Huggett-Aiyagari incomplete-market heterogeneous- agent framework, incorporating information asymmetry and a standard New Keynesian block. In the model, infinitely lived households receive idiosyncratic labor efficiency shocks, value leisure and consumption, and save and borrow by trading discount bonds with competitive financial intermediaries. Households can default on their loans. Fi- nancial intermediaries factor the default risk into the bond price, and endogenous credit limits arise. The core of the model mechanism lies in that credit is rationed due to adverse selec- tion, which is alleviated by accommodative monetary policy. Households differ in their default risk since they discount future default cost differently. But risk types are private information. Financial intermediaries thus cannot condition bond prices on households’ 1Detailed empirical analysis is in Section 2. 2 Personal Consumption Expenditure from NIPA Consumer Expenditure Survey, All Households 3 3 2 2 1 1 0 0 -1 -1 0 5 10 15 20 0 5 10 15 20 Households with Higher Default Risk Households with Lower Default Risk 3 3 2 2 1 1 0 0 -1 -1 0 5 10 15 20 0 5 10 15 20 Figure 1: Impulse Response of Consumption Notes: Units on the horizontal axis are quarters. Units on the vertical axis are percentage points. The solid lines are the impulse responses to a 100 basis point negative innovation in federal funds rate. The dashed lines are one standard deviation intervals constructed by bootstraps with 200 repetitions. Households’ default risk is measured based on risk premium charged on their auto loan interest rates. Monetary policy shocks are identified by ordering the federal funds rate last in a SVAR. The data in use is from 1984Q1to 2007Q4. risk types but can make inferences from their choices. In normal times when the inter- est rate is high, the risky type borrows more than the safe type since the former is less patient, borrowing thus signals a risky type, and credit is rationed. A temporary cut in the real interest rate encourages the safe type to increase borrowing relatively more than the risky type. The posterior probability of being a risky type conditional on borrowing decreases and credit is extended. To motivate an ex-ante measure of households’ default risk and associate the model with data, I assume financial intermediaries track a "credit score" of each household similar to Chatterjee et al. (2011), which denotes the prior probability the household is a safe type. I introduce unobservable preference shocks so that types cannot be revealed immediately in a single period. The credit score is thus used to form the expectation of default probability and to price bonds. The credit score is updated following Bayes rule 3 over time and reveals households’ types gradually. There are two important empirical regulations that I use to put quantitative discipline on the model. The first is the salient fact that consumers with higher default risk arealso those with higher marginal propensities to spend out of extended credit. The correlation between credit limit and marginal propensity to spend is crucial in determining how much of the extended credit is transformed into final aggregate demand. The second is the extent to which credit limit varies with credit score. This is indicative of the degree of adverse selection in the consumer credit market: if adverse selection is light and consumers’ behaviors perfectly reveal their types, prior information (credit score) should not be important in determining credit price and credit limit, and vice versa. I calibrate the model to match the cross-sectional distribution of default rate, credit limit, and marginal propensity to spend in the data. The main quantitative exercise is to study the transition path of the economy after monetary policy shocks modeled as unexpected shocks to the Taylor rule. I show that the model generates heterogeneous consumption responses qualitatively consistent with data. After a shock that lowers the nominal interest rate by 25 basis point on impact, the consumption response for the lower credit score group is 36% larger than the higher credit score group measured by percentage deviation on impact, or 63% larger measured by cumulated response through the transition. Heterogeneous consumption responses arise in the model because households differ in their marginal propensities to spend out of extended credit and because they face different responses of credit supply following the monetary policy shock. Credit sup- ply responds to the monetary policy shock for two reasons. First, given the type, the lower risk-free interest rate lowers the cost of rolling over debt and lowers the default rate. Second, the lower borrowing cost encourages the safe type to increase borrowing relatively more than the risky type, makes the pool of borrowers on average safer, and alters lenders’ beliefs on the risk types of borrowers. I examine the quantitative importance of different model mechanisms in explaining consumption responses through a series of counter-factual experiments. I show that the changes in lenders’ beliefs account for the majority of heterogeneity in consumption and credit supply responses. If the financial intermediaries were to ignore the changes in borrowing behaviors when making type inferences, the difference in consumption responses would be 31% lower measured by percentage deviation on impact, or 63% lower measured by cumulated response through the transition, and the difference in credit limit responses would virtually disappear. The changes in lenders’ beliefs are also quantitatively important in driving aggregate 4 consumption response. With the responses in lenders’ beliefs turned off, the aggregate consumption response is 20% lower on impact. I show that the effect of changes in credit supply on aggregate consumption is of similar magnitude as the effect of changes in the risk-free interest rate, and the latter is the key force in generating consumption response in traditional New-Keynesian models. The model makes sharp predictions on the distributional effects